The cassettes incorporate binding sites from two sources: sites that were evaluated experimentally (Lab Results), and sites identified algorithmically by applying machine learning to the experimental dataset (Deep Learning Model). The experiment, ML, and scoring are described in our publication.
Read more

Ranked: The algorithm allocates binding sites to cassettes according to their score rank. For N cassettes of M sites per cassette, the M top-scoring binding sites (scores with rank 1 to M) will appear in the 1st cassette. More generally, the binding sites with score rank of (M*(k-1)+1) to M*k will appear in cassette k. This method is recommended if only a few functional cassettes are needed, or if the user is interested in different levels of binding.

Balanced: The algorithm attempts to balance the scores of all generated cassettes. For N cassettes of M sites per cassette, the NxM top-scoring binding sites will be spread randomly among the N cassettes. This method is recommended if many cassettes with equal functionality are required for a given application. Due to randomness, each run may generate different output.